Skip to main content

· 6 min read
Jake W. Casselman
Annie Y.Y. Chang

In a world where technology has reshaped every facet of our lives, from how we communicate to how we navigate, it's surprising to find that our weather apps still trail behind, clinging to traditional methods that serve up raw data like temperatures and precipitation levels. This old-school approach misses the mark on what truly matters to us—the implications of these numbers on our daily activities and decisions. Whether you're a construction manager wondering if the weather will permit pouring concrete, or a parent planning a picnic, the question remains: what does 15°C and partly cloudy really mean for your day?

This gap between the detailed description of the physical environment and its actionable insights has persisted, requiring users to manually interpret how weather conditions might influence their plans. Addressing this challenge, ClimaLinks, a startup from ETH Zurich, is poised to enhance this process with an innovative approach to weather intelligence.

We are proud to unveil ClimaLinks' first innovation: the Weather Relationship Management (WRM) system. In reimagining our engagement with weather forecasts, ClimaLinks facilitates a seamless integration of weather data into meaningful, actionable insights tailored to the individual needs and scenarios of its users. Keep on reading and dive into the world of WRM with us, as we explore how this innovative system can transform your daily interactions with weather.

Understanding the Evolution of Weather Forecasting

Taking a step back to consider the historical context sheds light on why weather forecasts have traditionally been presented in a format that leaves much to the user's interpretation. Initially, weather information reached the public through newspapers and later, television news—mediums that, by necessity, catered to a general audience. This one-size-fits-all approach meant delivering raw data and leaving individuals to make sense of what that meant for their daily lives. It was a manual process, but within the constraints of the time, it worked.

With the advent of mobile technology, access to weather forecasts has dramatically expanded. Today, it's hard to find a smartphone without a weather app. Yet, this digital transition has mostly replicated the old models of presentation, focusing on visual enhancements and location flexibility but still providing data in its raw, uninterpreted form. This approach overlooks the transformative potential of digital platforms to provide contextual, actionable insights tailored to specific needs.

ClimaLinks is challenging this status quo by redefining the final mile of weather forecasting. ClimaLinks envisions a future where weather data isn't just accurate but is indispensable for forward-thinking strategies and operations. Our mission is to transform weather data from a mere reference to an active decision-making tool, enabling businesses to adopt a proactive approach.

No longer is weather information a mere icon on your screen indicating it's sunny and 16°C. ClimaLinks integrates weather insights directly into your planning tools, making it an indispensable part of decision-making processes rather than an afterthought.

A Glimpse into the Future: Unveiling the Power of AI in Weather Forecasting From static print to dynamic digital: The transition of weather forecasting through the ages, yet the essence remains unchanged—raw data awaiting personal interpretation

Beyond Precision: Marrying Accuracy with Usability in Weather Forecasting

It should still, however, be emphasized that the weather forecasting industry has witnessed impressive innovation, particularly in improving forecast accuracy. Teams from world-renowned universities and major agencies like NOAA, and ECMWF have made significant strides, and recently, private sector giants like Google, NVIDIA, and Microsoft have joined the fray with their cutting-edge AI models (check out our previous blog). These models, such as Google's Graphcast and NVIDIA's FourCastNet, are competing to deliver forecasts that are not just accurate but incredibly rapid, producing results thousands of times faster than traditional methods.

Yet, this rapid innovation has largely centered on one half of the forecasting coin: accuracy. The other side, usability, remains an area less trodden. Historically, when the bottleneck was forecast accuracy, this focus was justified. But now, as we find ourselves awash in high-quality forecasts that can reliably predict weather days in advance, we must address the other challenge. It's no longer just about having accurate data; it's about making that data work for us in the most efficient and practical ways possible.

ClimaLinks stands at the vanguard of this new frontier, ensuring that usability receives as much attention as accuracy. By harnessing the power of AI, ClimaLinks is not simply providing data; it's offering a suite of tools that integrate seamlessly into users' lives, providing them with personalized, actionable weather insights. This is the next step in the evolution of weather forecasting, where data becomes a cornerstone of daily decision-making, moving from a passive information source to an active component of our daily lives and businesses.

WRM Dashboard Discover the Dashboard: A first look at WRM’s user-centric interface, designed to bring weather intelligence to your fingertips.

At ClimaLinks, we proudly present our first major innovation: the Weather Relations Management (WRM) platform and API. Envisioned as both a standalone dashboard and an API for existing SaaS management tools. Our system is not just an add-on; it's a gateway to a sophisticated weather insights not previously possible.

We have tailored WRM to be responsive to our users' needs, contextualizing tasks, assets, and scenarios into hyper-granular indices. Whether planning a barbecue at the beach, cleaning high-rise windows, or laying concrete foundations, our engine adapts to any task. Equipped with a built-in AI Agent, WRM constructs your personalized weather indices, transforming complex meteorological data into clear, actionable insights.

Welcome to a new era where dealing directly with raw weather data becomes a thing of the past. ClimaLinks' WRM platform empowers you to optimize operations with our Task Monitor, safeguard assets with proactive risk assessments, and preemptively address potential vulnerabilities using our Asset Monitor. Say goodbye to the days of being caught off guard by the weather. With our Schedule Tracker, you can plan ahead with confidence, knowing that our system continuously analyzes forecast updates, ensuring you're never left unprepared or unexpectedly disadvantaged.

The future is here, and with ClimaLinks' WRM, it's tailored, proactive, and ready to seamlessly integrate into the fabric of your daily operations. Weather forecasting is no longer just about prediction—it's about intelligent preparation and strategic advantage.

We are just at the beginning of this journey, and the potential for growth and innovation is boundless. As ClimaLinks continues to evolve and expand the capabilities of WRM, we invite you to be part of this revolution. Stay tuned for further developments, or better yet, step into the future with us—experience the power of WRM firsthand.


For a demo or to learn more about how WRM can be integrated into your life or business, contact us. Join us as we chart a new course in weather forecasting, one where every forecast is a step towards smarter decisions and a more resilient tomorrow.

· 10 min read
Jake W. Casselman

A Glimpse into the Future: Unveiling the Power of AI in Weather Forecasting Mapping the Future of Meteorology: AI-driven Innovations by Google, NVIDIA, Microsoft, and Huawei Redefine Weather Prediction (AI-generated Image by OpenAI)

You might have recently seen that Google is making strides in AI-driven weather forecasting with their Graphcast model. But did you know that many other large tech companies are also entering the scene? In this article, we dive into the advancements in AI meteorology innovations from the private sector, putting a spotlight on key players like NVIDIA's FourCastNet, Microsoft's ClimaX, and Huawei's Pangu-Weather and exploring how these technologies have the potential to revolutionize the field of meteorological science.

🎯 AI Methods in Weather Forecasting: Same Goal, Different Approaches

The quest for the most effective AI method in weather forecasting is a dynamic and ongoing challenge. While Google has been one of the frontrunners with models like Graphcast and Met-Net-3, the industry is witnessing a variety of innovative approaches by other tech giants.

Google's Graphcast and Met-Net-3

Graphcast excels in providing rapid and accurate weather predictions. This model distinguishes itself with forecasts that extend up to 10 days into the future, demonstrating a resolution accuracy of about 28x28 km at the equator. While Graphcast is adept at short to medium-range forecasts, it encounters challenges in enhancing resolution and sustaining accuracy for longer forecast periods. In contrast, Met-Net-3, another innovative model from the same team, specializes in high-resolution forecasting. This model is capable of delivering forecasts with an impressive 1x1 km resolution, offering updates every few minutes. Its forte lies in its precision, although it primarily focuses on short-term forecasting, with a lead time of just 24 hours. For in-depth insights, visit Google's official Graphcast blog and Met-Net-3 blog. For the scientific paper on Graphcast, see here.

Technical Reading on Google's Graph Neural Network (GNN)
  • Approach: Google's GraphCast, developed by DeepMind, uses Graph Neural Networks (GNNs) for weather forecasting, focusing on spatially structured data.

  • Notable Strengths: It provides rapid and precise weather predictions, while being a relatively small model (36.7 million parameters). This is significant as a 10-day forecast can be run in under a minute on relatively basic infrastructure (in their study, they use a TPU v4 device).

  • Notable Limitations: Because the training data is ERA5 with a resolution of 0.25 degrees, this may produce challenges when trying to increase the resolution (such as to sub-km). Uncertainty is also handled by "spatially blurring" the forecasts, which becomes an important constraint at longer lead times (e.g., forecasting several days into the future).

NVIDIA's FourCastNet

FourCastNet's strength lies in its exceptional speed, processing forecasts much faster than traditional models. Its current limitation is in sometimes producing results that deviate from established physical laws, a gap NVIDIA aims to bridge with future enhancements. Explore more on NVIDIA's official blog. For the scientific paper on FourCastNet, see here.

Technical Reading on NVIDIA's Fourier-based neural network
  • Approach: NVIDIA's model uses a Fourier-based neural network framework and runs on cuDNN-accelerated TensorFlow. It operates on NVIDIA GPUs, achieving high-speed computation (A100 GPU, similar to Chat GPT).

  • Notable Strengths: It can process extensive ensemble forecasts rapidly, making it suitable for real-time applications. To put this into context, NVIDIA claims that they are "45,000 times faster than traditional NWP models" (source). This is especially useful when considering ensemble forecasts and detection of extreme events. They also expect to eventually train the FourCastNet to "predict weather on sub-km scales."

  • Notable Limitations: Like many other purely data driven AI weather models, the output may not always strictly obey known physical laws (read: produces unrealist results). Noting this in their research paper, NVIDIA plans to incorporate "Physics-informed Machine Learning" methods in the future.

Microsoft's ClimaX

ClimaX handles various datasets effectively, particularly in downscaling for local forecasts. It uses separate models for different forecast intervals, a limitation addressed by ClimaX-iter which allows iterative predictions, though this can affect long-term forecast accuracy. Learn more about their approach on Microsoft's ClimaX overview and access the scientific paper here.

Technical Reading on Microsoft's Vision Transformers (ViT)
  • Approach: Microsoft's ClimaX employs a multi-dimensional image-to-image translation architecture based on Vision Transformers (ViT).

  • Notable Strengths: It's designed to handle a wide variety of weather and climate modeling training data ("heterogeneous datasets spanning different variables, spatio-temporal coverage, and physical groundings"). Using a ViT method also allows for training the model on spatially-incomplete datasets. Notably, this method also works well for being able to downscale data to get more localized results.

  • Notable Limitations: In ClimaX, each model instance is trained for a specific lead time, necessitating separate models for different forecasting intervals. ClimaX-iter addresses this by enabling iterative predictions over multiple lead times with a single model. This approach offers greater efficiency and scalability, but it compromises accuracy in longer-term forecasts due to error accumulation.

Huawei's Pangu-Weather

Pangu-Weather stands out as one of the few models implementing a 3D Earth Specific Transformer architecture, a notable advancement in AI weather forecasting. This 3D approach is key to its accuracy, allowing the model to capture intricate atmospheric relationships. Its rapid forecasting capabilities are on par with NVIDIA's FourCastNet. However, the complexity of 3D data processing raises computational demands, which poses challenges for scalability and practical application. This mirrors the balance seen in advanced AI models like OpenAI's ChatGPT, where increased sophistication often comes with higher resource requirements. Learn more about their approach on Pangu-Weather's official page and access the scientific paper here.

Technical Reading on Pangu-Weather and Huawei's 3D Earth Specific Transformer (3DEST)
  • Approach: This model uses a 3D Earth-Specific Transformer (3DEST) architecture to process complex 3D meteorological data.

  • Notable Strengths: Pangu-Weather can produce forecasts extremely rapidly, suggesting that they are similar to NVIDIA's FourCastNet (i.e., 45,000 times faster than traditional numerical models). This means that forecasts can be done in "real-time" (i.e., second(s)). They also use a 3D deep neural network, which can be more accurate than 2D networks (i.e., NVIDIA's FourCastNet), as it can capture intrinsic relationship between different atmospheric levels.

  • Notable Limitations: The primary challenge for the Pangu-Weather system, akin to advanced AI models like OpenAI's ChatGPT, lies in the trade-off between model sophistication and computational demands. As the system incorporates more complex 3D data and hierarchical algorithms for accuracy, it faces increased computational overhead, particularly in memory usage, which restricts the scalability and practical application of the technology.

🌐 Broadening the Horizon: Beyond the Leading Tech Giants

This article offers a snapshot of the private sector's advancements in meteorological innovation. There are also many other private companies like Jua (Swiss Startup) and Meteomatics (Swiss Company) at the forefront, with Jua developing AI models and Meteomatics using meteodrones for unique weather insights.

The European Centre for Medium-Range Weather Forecasts (ECMWF) plays an indispensable role that extends beyond the private sector. Their models are benchmarks in the field, often used by private companies for comparison. Furthermore, ECMWF, in collaboration with Copernicus, provides the ERA5 reanalysis model, which is a crucial training dataset for many of these emerging AI models. Demonstrating their support for innovation, ECMWF actively embraces the private industry by hosting groundbreaking models like Google's Graphcast, NVIDIA's FourCastNet, and Huawei's Pangu-Weather on their platform. For an in-depth look, visit ECMWF's AI showcase.

ECMWF has also launched their own AI model, the AIFS (Artificial Intelligence/Integrated Forecasting System), enhancing their Integrated Forecasting System (IFS). Live model runs of AIFS can be seen at ECMWF's AIFS charts. Learn more from their blog: ECMWF unveils alpha version of new ML model.

The evolving field of AI in weather forecasting, driven by private companies, academia, and inter-governmental collaborations, promises more precise and efficient meteorological predictions.

🧩 Practical Challenges: The Initial Conditions Hurdle

While a growing number of cutting-edge AI weather models are becoming open-source and accessible on platforms like GitHub, a significant challenge persists for average users: the acquisition of initial conditions. These conditions, representing the current observed state of the weather, are essential for initiating any forecast model. This situation highlights that, despite the technical advancements and availability of sophisticated models, practical application still faces substantial hurdles, especially in terms of data accessibility. Addressing these challenges, innovative companies like Jua and Meteomatics are exploring novel solutions, including leveraging IoT and drones to actively gather atmospheric data.

1. Ensembles and Rapid Forecasting

  • Ensemble Forecasts: With the acceleration of forecast capabilities in models like Huawei's Pangu-Weather and NVIDIA's FourCastNet, the adoption of large ensemble forecasts may increase. These ensembles are crucial for better predicting extreme weather events and for extending forecast ranges.
  • Real-Time Forecasting: Advancements in forecasting speed could lead to more frequent updates, potentially evolving towards continuous forecasting. This change highlights the importance of efficient data pipelines for integrating initial conditions. Insights from nowcasting could guide the development of this new forecasting paradigm.

2. Democratization of Forecasting

  • Lightweight Models: Google's lightweight and fast models have the potential to democratize weather forecasting, particularly in less developed regions. However, this trend might be overshadowed by the movement towards open weather data policies by large inter-governmental institutions. With organizations like ECMWF, NOAA, and soon MeteoSwiss embracing open data policies, a global shift towards more accessible weather forecasting tools is likely.

At ClimaLinks, we find ourselves at the forefront of a transformative era in weather forecasting. We're not just witnessing the AI revolution in meteorology; we're an integral part of it. As accurate weather forecasts become more accessible and commonplace, we recognize that simply having forecasts isn't enough. The true game-changer lies in how we utilize this expanding pool of accurate data.

This is where ClimaLinks takes center stage. We're innovating the first-ever Weather Relationship Management (WRM) platform. Our aim? To revolutionize how weather data is used. As weather forecasts become more freely available and also precise, we're helping to shift the industry's focus. It's not just about predicting the weather anymore; it's about understanding and using this information to its fullest potential.

Our WRM tool is specifically designed to tackle this 'last-mile' challenge in weather forecasting. We're committed to helping our users not just access the best weather information but truly leverage it. At ClimaLinks, we envision a world where weather data becomes an indispensable tool for proactive decision-making, transforming the way individuals, businesses, and governments respond to meteorological changes.

In this new era, where the abundance of data is both an opportunity and a challenge, ClimaLinks is leading the way in making weather data a cornerstone of strategy and operations.

Take Action

Ready to take your weather strategy to the next level? At ClimaLinks, we're more than just forecasters; we're your partners in weather intelligence. Whether you're a business looking to optimize operations, a government agency aiming to enhance public safety, or just someone who wants to stay ahead of the weather, ClimaLinks has the tools and expertise you need.

Try our products:

· 3 min read
Jake W. Casselman
Annie Y.Y. Chang

Welcome to the official blog of ClimaLinks! We're thrilled to have you here. In this post, we'll dive deep into what we do, the problems we're addressing, and how our innovative solutions are shaping the future.

At ClimaLinks, we offer top-tier weather intelligence software solutions, transforming mere weather data into actionable insights. Our aim? To redefine how businesses handle weather-related opportunities and challenges.

Our Initial Areas of Focus:
  • Core: Construction and Civil Engineering
  • Pilots in: Facility Management, Agriculture

We are on a quest to assist industries vulnerable to weather disruptions. Our offerings include our flagship ClimaLinks Weather Relations Management (WRM) Software, as well as ClimaLinks Data-as-a-Service.

❓ The Big Challenge

Despite the advancements in forecasting accuracy, many businesses struggle to convert this precision into actionable strategies. This gap can result in unforeseen costs like project delays, missed opportunities, and resource mismanagement.

In the construction industry, delays are all too common due to a myriad of factors. However, a staggering 45% of these delays are attributed solely to weather conditions. This underscores the pressing need not just for precise but also actionable weather insights.

The Gist

Traditional weather platforms tend to offer raw data without much contextual interpretation. While this might be manageable for individuals, companies, given their intricate operations, struggle to proactively leverage this information. Without tailored insights, businesses miss out on opportunities for proactive decision-making that would otherwise be straightforward.

But that's where we step in. Our mission is to bridge this "final mile" of innovation, turning abundant weather data into tangible insights. With the rise of Generative AI, we're poised to address this gap and usher in a new era of weather intelligence.

🌍 Our Vision and Mission

While many view weather forecasts as mere passive indicators, we at ClimaLinks see them as opportunities. Our mission is simple yet transformative: turn weather data from a passive reference into an active decision-making tool.

We envision a future where weather data isn't just accurate, but indispensable for proactive strategies and operations.

🎯 Company Goals and Objectives

Our flagship product, the Weather Relations Management (WRM) software, is set to revolutionize how businesses interpret weather forecasts. Our primary market is Switzerland, but we have plans to expand to Canada and Taiwan.

We're not just about data. As we evolve, the integration of IoT will further enrich our data sources, providing unprecedented granularity and real-time updates. We're all about innovation and adaptability, and we're geared up to set a new benchmark in weather intelligence.

💡 Innovation and Advancements

Our WRM software introduces a fresh approach to weather intelligence, catering specifically to B2B applications. The platform is versatile and moulds itself to different industries' unique requirements. Plus, businesses can integrate ClimaLinks WRM into their existing systems via Power Automate or our API, ensuring seamless decision-making based on our insights.

Now What?
  • Stay Informed: Sign up now for our newsletter! Sign Up Form.
  • API products: Explore the docs at your own pace! API Docs.
  • Try our products: Reach out to us to get a demo! Send us an email.